Update noRag.py

noRag.py

@@ -1,6 +1,7 @@
 # noRag.py
 
-from fastapi import APIRouter, HTTPException
+import uuid
+from fastapi import APIRouter
 from pydantic import BaseModel
 from groq import Groq
 from pymongo import MongoClient
@@ -8,35 +9,45 @@ from config import CONNECTION_STRING, CHATGROQ_API_KEY, CUSTOM_PROMPT
 
 router = APIRouter(prefix="/norag", tags=["noRag"])
 
-# Initialize Groq client and MongoDB
+# clients
 client = Groq(api_key=CHATGROQ_API_KEY)
-mongo  = MongoClient(CONNECTION_STRING)
-db     = mongo["edulearnai"]
-chats  = db["chats"]
+mongo = MongoClient(CONNECTION_STRING)
+db = mongo["edulearnai"]
+chats = db["chats"]
 
 SYSTEM_PROMPT = "You are a helpful assistant which helps people in their tasks."
 
+
 class ChatRequest(BaseModel):
     session_id: str
-    question:   str
+    question: str
+
+
+@router.post("/session", summary="Create a new chat session")
+async def create_session():
+    session_id = str(uuid.uuid4())
+    chats.insert_one({"session_id": session_id, "history": [], "summary": ""})
+    return {"session_id": session_id}
+
 
-@router.post("/chat", summary="Ask a question to the noRag assistant")
+@router.post("/chat", summary="Send a question to the assistant")
 async def chat_endpoint(req: ChatRequest):
-    # Fetch or create session in MongoDB
     doc = chats.find_one({"session_id": req.session_id})
     if not doc:
+        # if session not exist, create it
         doc = {"session_id": req.session_id, "history": [], "summary": ""}
         chats.insert_one(doc)
 
     history, summary = doc["history"], doc["summary"]
 
-    # Summarize if history too long
+    # auto-summarize if too many turns
     if len(history) >= 10:
         msgs = [f"{m['role']}: {m['content']}" for m in history]
         combined = summary + "\n" + "\n".join(msgs)
         sum_prompt = (
             "Summarize the following chat history in one or two short sentences:\n\n"
-            f"{combined}\n\nSummary:"
+            + combined
+            + "\n\nSummary:"
         )
         sum_resp = client.chat.completions.create(
             model="meta-llama/llama-4-scout-17b-16e-instruct",
@@ -49,15 +60,15 @@ async def chat_endpoint(req: ChatRequest):
         summary = sum_resp.choices[0].message.content.strip()
         history = []
 
-    # Build the prompt for Groq
-    chat_hist_text = "\n".join(f"{m['role']}: {m['content']}" for m in history)
+    # build full prompt
+    hist_text = "\n".join(f"{m['role']}: {m['content']}" for m in history)
     full_prompt = CUSTOM_PROMPT.format(
         context=SYSTEM_PROMPT,
-        chat_history=chat_hist_text or "(no prior messages)",
-        question=req.question
+        chat_history=hist_text or "(no prior messages)",
+        question=req.question,
     )
 
-    # Get the assistant’s answer
+    # get answer
     resp = client.chat.completions.create(
         model="meta-llama/llama-4-scout-17b-16e-instruct",
         messages=[{"role": "user", "content": full_prompt}],
@@ -68,13 +79,13 @@ async def chat_endpoint(req: ChatRequest):
     )
     answer = resp.choices[0].message.content.strip()
 
-    # Persist the Q&A
-    history.append({"role": "user",      "content": req.question})
+    # persist
+    history.append({"role": "user", "content": req.question})
     history.append({"role": "assistant", "content": answer})
     chats.replace_one(
         {"session_id": req.session_id},
         {"session_id": req.session_id, "history": history, "summary": summary},
-        upsert=True
+        upsert=True,
     )
 
     return {"session_id": req.session_id, "answer": answer, "summary": summary}